Rin Arakaki

Product Manager, Technical Lead, Voice Team, Algoage

Nov 2025 — Current, Tokyo, Japan

Real-time AI copilot for contact-centre operators — telephony, transcription, knowledge and topic detection as one system on the customer's own stack.

• Founded a real-time AI copilot for contact-centre operators and led it from first commit to multi-tenant production in nine months, now running across seven customers.
• Grounded the product in observed practice — spent time on-site at a customer's contact centre learning how operators actually work, and ran user interviews to test the product hypothesis before committing to a roadmap.
• Owned the architecture across telephony integration, real-time voice, AI agent, multi-tenant backend, browser client and infrastructure, with service contracts defined before implementation and business rules held as data rather than code.
• Designed the dimension model the product organises itself around — a customer's own classification axes, resolved by one embedding-based lookup serving both knowledge management and live topic detection.
• Made knowledge search grounded by construction — doc2query expansion generating each document's answerable queries, verified offline, so an operator is never offered a search that returns nothing and the serve path does no retrieval at all.
• Closed the knowledge loop — operator flags, usage signals and post-call AI gap detection converging on one queue, with edits authored and approved separately and published live to the next call.
• Built the platform that measures whether the AI is any good — eleven dimensions scored against synthetic calls driven through the real telephony path, with operator accept-versus-edit rate as ground truth.
• Identified a recurring reliability failure class the team had been patching case by case, and replaced six rounds of defensive fixes with a single design invariant.
• Established reliability as a governed practice: service-level objectives owned as code, burn-rate alerting, and an error-budget policy binding budget exhaustion to a deploy freeze.
• Grew the team from two to fourteen — five engineers, three project managers, an engineering manager, three QA and a designer — and defined a new Product Ops function after recognising work chartered as project management as a structurally different role.
• Won seven customers on product strength — exhibition demand, and a vendor referring its own customers after evaluating the copilot — with integration breadth across their existing stacks carrying as much of the value as the AI.
• Built the agent-operated pipeline — executable specifications, automated code review, documentation reconciliation — that produced and maintains a million-line codebase without a line hand-written, with correctness held by invariant tests that fail the build and a test suite a third the size of the application code.

Technologies — Amazon Connect, Twilio, LiveKit, WebRTC, Deepgram, Cartesia, Anthropic Claude, OpenAI, Gemini, Chrome Extension, Model Context Protocol, Spec Kit, Temporal, OpenTelemetry, pgvector, Amazon Web Services, Terraform

Software Engineer, Part-Time, Archemis

Oct 2025 — Current, Tokyo, Japan

AI-native crypto trading hedge fund.

• Owned a live trading strategy end to end in Rust — six macro-economic indicators resolved into a correlation-weighted signal sizing a Bitcoin futures position — through to production deployment.
• Wrote the trading logic and data-validation specifications, then the input validation they define — non-finite values, negative yields, market-holiday gaps, insufficient history.
• Specified the next strategy before implementing it — auditing the analyst's definition and documenting its contradictions, with every threshold stated as inclusive or exclusive.

Technologies — Rust, Google Cloud Platform, Kubernetes, Terraform

Software Engineer, aisaac

Jul 2025 — Nov 2025, Tokyo, Japan

Dating app for over-40s in Japan.

• Architected and implemented web payments end to end — cross-platform authentication, subscription lifecycle, user-behaviour tracking, purchase event logs, refunds and coupons, and a double-charge race fix — improving ROAS by 30%.
• Built revenue forecasting over the transactional data to support business planning.
• Researched and designed a new matching algorithm combining TF-IDF and vector calculation.

Technologies — SQL, TypeScript, Google Cloud Platform, React Native, Google Analytics, Adjust

Forward Deployed Engineer, JAPAN AI

Apr 2025 — Jun 2025, Tokyo, Japan

Customer deployments of the AI call centre product.

• Built and deployed operator-customer matching for a call-centre client — solved with constraint programming, and a CSV upload path so the client could supply its own data.

Technologies — OR-Tools, C++, Python, Google Cloud Platform

Full-Stack Engineer, Speech Team, JAPAN AI

Oct 2024 — Apr 2025, Tokyo, Japan

LLM and generative AI research and development.

• Architected and led an AI call centre platform — infrastructure, schema, UI and agent design — researching UDP-based stateful communication protocols and the Kubernetes network model. Presented a working demo to the CEO, and the platform shipped to three customers.
• Built the real-time speech-to-speech stack — a VoIP client speaking the telephony protocols directly with self-hosted transcription, then migrated onto a managed real-time agent framework with pluggable speech synthesis, egress, hangup and interruption handling — and the control-plane API dispatching agents into rooms and recording call artefacts.

Technologies — LiveKit, Twilio, SIP, WebRTC, Whisper, Kubernetes, gRPC, Python, Google Cloud Platform, RunPod

Full-Stack Engineer, Kakeai

Sep 2023 — Sep 2024, Tokyo, Japan

One-on-one meeting platform with real-time AI transcription and summarisation.

• Re-architected the core calendar and scheduling platform — external sync, recurrence rules, third-person event creation, preferred time slots — replacing it with a third of the code while adding features, then built mentor-mentee pairing on it with a stable matching algorithm.
• Ran a causal study of one-on-one practice against engagement outcomes — six survey waves and 7,400 one-on-one records — estimating treatment effects with propensity-score matching, difference-in-differences and instrumental variables.

Technologies — Python, TypeScript, Amazon Web Services, Terraform, Ruby on Rails

Lead Data Engineer, Viagate

Apr 2023 — Aug 2023, Tokyo, Japan

Marketing research platform using eye tracking and facial expression recognition.

• Built the data platform from scratch — event-driven ingestion into a self-hosted time-series store, columnar compaction, aggregation and querying — with the whole estate defined as code.
• Defined the data layer interfaces-first, introducing a single binary schema contract across iOS and web clients, and led a backend team of three.
• Held SQL to the same standard as application code: definitions and queries under a linter, with a test suite and query benchmarks.
• Implemented the video and image processing pipeline behind the platform's data visualisation.

Technologies — OpenCV, Apache Arrow/Parquet, Protocol Buffers, PostgreSQL, InfluxDB, Python, Amazon Web Services

Technical Lead, Operations Lead, Delipicks

Jun 2019 — May 2020, Tokyo, Japan

B2B food delivery startup; joined 6 months after launch.

• Took charge of daily operations and defined the system requirements from inside them with the CEO — business processes, system and the standard operating manuals as one specification.
• Built the real-time system connecting customer orders, restaurant menu planning and delivery, removing the operations and support headcount the business model would otherwise have needed to scale.
• Made menu and route planning tractable with constraint programming — the bottleneck on growth — and automated the weekly planning meeting out of existence.
• Modelled delivery routes rather than service areas as the unit of the operation, collapsing daily menu registration to a single entry per route and recovering thirty minutes a day.
• Ran UX workshops and over ten user interviews, prototyping the operator interface and briefing designers from the operations floor on how mistakes actually happen.
• Assembled a cross-functional team of five, recruiting three and sourcing the mathematical optimisation specialist the approach required.
• Worked both sides of the marketplace — cold calling the companies that would buy, opening the restaurant relationships that would supply them, deploying the operations app into supplier workplaces, then negotiating deliverables.
• Sold on the street and at rental offices alongside the CEO, and handed brochures at office doors in the mornings, to find the LTV and CAC the business model had to prove.
• Instrumented the operation with a sales-data system, then negotiated supplier pricing and commercial terms from it, including compensation for delivery errors the business had previously absorbed.

Technologies — OR-Tools, Firebase, Python, JavaScript, Google Cloud Platform

Software Engineer, Styler

Aug 2018 — Mar 2019, Tokyo, Japan

Fashion retail platform bridging online and in-store experience (OMO — online-merge-offline).

• Owned infrastructure selection, database design and API development, and led the move from a monolith to a single-page front end — delivered through others, assigning execution, following to completion.
• Bridged an engineering team of six nationalities working in English to design, business development and media working in Japanese, as development came in-house from offshore.

Technologies — Kubernetes, Python, Google Cloud Platform, Ruby on Rails

Software Engineer, MarketEnterprise

Apr 2016 — Dec 2017, Tokyo, Japan

Resale and rental business with complex internal operations.

• Spent three months working every stage of the operation — quoting used-goods prices in the call centre, photographing, describing and listing stock to e-commerce and auction channels in the fulfilment centre, collecting goods in delivery logistics — then designed the internal systems as someone who had used them.
• Modernised long-established internal systems, opening the move to microservices with the organisation-wide authorisation API. Took it over from a senior engineer after it stalled at 7 seconds and neared cancellation — investigated the specification until certain of the design, then proved authorisation could resolve in two queries at 0.2s.
• Owned the purchase-data analysis behind the company's public relations — working with the fulfilment centre head and public relations to define what data could be acquired, then correcting and normalising the source and producing category rankings and trends that were published and serialised as an ongoing column.

Technologies — MySQL, Python, JavaScript, Amazon Web Services

The Open University of Japan, Bachelor's Degree, Liberal Arts

Apr 2012 — Mar 2016, Tokyo, Japan

• Completed degree requirements in 2 years.

Founder, Synthetic

Dec 2020 — Present, Tokyo, Japan

Independent research initiative — formal AI reasoning and automated proof generation. Self-funded and self-directed since 2020; sustained across 5,000+ commits.

• Designing a graph and network-based search and indexing system for mathematical formulae, applied to automated theorem discovery.
• Investigating the theorem provers — Lean, Coq, Isabelle — and their proof search engines.
• Researching formal proof generation through LLM and reinforcement-learning feedback loops.

Technologies — Rust, Python, TypeScript